LLM-Entary, My Dear Watson
This input is how you tell the robot exactly what you need it to do.
What is a prompt?
This model comes in different "sizes" like Pro for complex tasks and Flash for when you need a lightning-fast response.
What is Gemini?
This is the part of a prompt that tells the AI who it is supposed to be, such as a "Golden Retriever" or a "No-Nonsense Drill Sergeant."
What is a persona/role?
You can upload up to 50 different government reports or policy memos to this tool and "chat" with them as if they were a single, highly-informed expert.
What is NotebookLM?
This is a common "ghost" occurring when an AI confidently tells you that George Washington invented the Internet.
What is a hallucination?
This tech describes a machine’s ability to create original content that has never existed before.
What is Generative AI?
You bought a "Simple" bookshelf that came with an instruction video and a 20-page PDF booklet; this word describes an AI that can watch the video and read the text booklet at the same time to explain Step 4 to you in plain English.
What is multimodal?
Instead of just asking for a "funny joke," you give the AI three examples of your favorite puns first; this "Monkey See, Monkey Do" technique is called this.
What is few-shot prompting?
This is the tool that allows travel agencies to automatically categorize thousands of user-uploaded photos based on the location and subject sentiment, like "Happy teenagers in front of the Eiffel Tower."
What is Cloud Vision API?
This occurs when a facial recognition system works 99% of the time for one group of people but fails frequently for another because its training data only has photos from one ethnic group.
What is bias?
This is what the "LLM" in LLM-entary actually stands for.
What is Large Language Model?
This is the "Digital Brain" of AI, modeled after the human nervous system to help computers learn patterns instead of just following a list of rules.
What is a neural network?
You are using this technique when you simply say "Write a poem about a Sacramento summer" and hope the AI just "gets it" on the first try.
What is zero-shot prompting?
If your agency is drowning in thousands of handwritten permit applications, this is the specialized "Processor" that can read the handwriting and turn it into a clean digital spreadsheet.
What is Document AI?
This happens when an AI that was trained on 2023 fashion trends starts giving bad advice because it doesn't know what’s popular in 2026.
What is data drift?
This is the "Currency" of the AI world. Every time you send a prompt, you are "spending" a certain amount of these to get an answer back.
What is a token?
This is the AI's "Creative Thermostat," the setting you turn up if you want a wild, imaginative poem and turn down if you want a dry, factual report.
What is temperature?
This is the AI version of "Show Your Work," telling the robot to explain its reasoning step-by-step instead of just blurting out a final (and possibly wrong) answer.
What is Chain-of-Thought prompting?
You are a developer who needs to update an old piece of software and want an AI to explain what the original code is actually doing.
What is Code Assist?
This occurs when an AI "memorizes" its training data so perfectly that it fails to work when it sees a brand-new, real-world example.
What is overfitting?
This term describes messy data, including emails and videos, that does not fit into a standard database.
What is unstructured data?
This is the AI's "Anchor," a technique that forces the model to look at your specific documents (like PDFs or websites) instead of just "guessing" from its imagination.
What is grounding?
While it sounds like a job for a software developer, this is the specific practice of carefully crafting, testing, and refining the instructions given to an AI to ensure the most accurate and useful output.
What is prompt engineering?
You have a decade’s worth of historical tax data and need to predict which neighborhoods will see the most growth next year. This is your AI "Fortune Teller."
What is BigQuery?
This phrase describes the essential safety practice where a person must review and approve an AI’s output.
What is Human-in-the-loop?